A value far from the rest of a collection can be an error, an unusual but real event, or a clue that several different groups have been mixed. Its appearance alone does not settle the explanation.
Inspect the original record and the method that produced it. Check units, entry rules, and surrounding observations. Avoid removing a value solely because the chart becomes tidier without it.
If a value is excluded for a justified reason, record the rule and consider how the result changes with it included. An unusual observation becomes useful when the investigation is transparent enough for someone else to understand the decision.
- Inspect the source and measurement method.
- Check units and entry rules.
- Document any exclusion and its effect.
Picture this situation.
A single long task duration could include an interruption. Investigating the source can distinguish a useful exception from an incorrectly recorded observation.
A second way to look.
Look for the comparison behind the visual impression. The baseline, time window, and group definition can change what a pattern seems to mean.
Follow a related question
Distinguish an amount from a rate.
Energy and power answer different questionsDistinguish model training from inference.
Training and use are different stagesKeep learning
Related background to continue exploring this subject.
W3C: a primer for tabular data RFC Editor: the CSV format